Aqueous two-phase system patterning ofdetection antibody solutions forcross-reaction-free multiplex ELISAJohn P. Frampton1, Joshua B. White1, Arlyne B. Simon2, Michael Tsuei1, Sophie Paczesny3

& Shuichi Takayama1,2

1Department of Biomedical Engineering, University of Michigan, 2Department of Macromolecular Science and Engineering,University of Michigan, 3Department of Pediatrics, Indiana University School of Medicine.

Accurate disease diagnosis, patient stratification and biomarker validation require the analysis of multiplebiomarkers. This paper describes cross-reactivity-free multiplexing of enzyme-linked immunosorbentassays (ELISAs) using aqueous two-phase systems (ATPSs) to confine detection antibodies at specificlocations in fully aqueous environments. Antibody cross-reactions are eliminated because the detectionantibody solutions are co-localized only to corresponding surface-immobilized capture antibody spots. Thismultiplexing technique is validated using plasma samples from allogeneic bone marrow recipients. Patientswith acute graft versus host disease (GVHD), a common and serious condition associated with allogeneicbone marrow transplantation, display higher mean concentrations for four multiplexed biomarkers (HGF,elafin, ST2 and TNFR1) relative to healthy donors and transplant patients without GVHD. The antibodyco-localization capability of this technology is particularly useful when using inherently cross-reactivereagents such as polyclonal antibodies, although monoclonal antibody cross-reactivity can also be reduced.Because ATPS-ELISA adapts readily available antibody reagents, plate materials and detection instruments,it should be easily transferable into other research and clinical settings.

The enzyme-linked immunosorbent assay (ELISA) is used in clinical and laboratory settings to accurately andreproducibly quantify soluble proteins1. ELISA is typically performed in a sandwich format by immobilizingcapture antibodies on an assay plate, allowing specific antigens to associate with the surface immobilized

antibodies, and then detecting the antigens by way of detection antibodies that generate chromogenic, fluorescentor chemiluminescent signals2,3.

Owing to their versatility and reliability, ELISAs have been used to detect HIV/AIDS4, malaria5, cancer6,7 andinflammatory/autoimmune diseases8,9, among numerous other pathologies. However, conventional singleplexELISA formats are limited by high reagent costs, inefficient use of patient samples and an inability to preventantibody cross-reactions when multiplexed10,11. For example, multiplex platforms that spatially segregate captureantibodies to many individual spots or beads within an assay-well can greatly increase ELISA throughput;however, this advantage is often undercut by problems associated with cross-reactions among detection anti-bodies applied as a cocktail12,13. The interactions among detection antibodies or between detection antibodies andinappropriate capture antibodies or antigens can lead to false-positive or false-negative readouts. The chance ofsome form of cross-reactivity greatly increases with each new detection antibody added to a multiplex panel,especially when the antibodies are polyclonal, as is the case for the majority of commercially available ELISA kits.

Many diseases, for example acute graft versus host disease (GVHD), cannot be diagnosed with sufficientspecificity and sensitivity using single biomarkers14. Acute GVHD occurs in approximately half of allogeneicbone marrow transplantation recipients when donor immune cells recognize the host tissues as foreign and attackthem. This reaction can be minimized by carefully matching the donor and host tissues and using prophylacticimmunosuppression, but it is still the leading cause of non-relapse mortality in this population. Pre-transplantclinical or transplant characteristics have minimal ability to predict acute GVHD outcomes. Currently, acuteGVHD is diagnosed by clinical symptoms in three organ systems (skin, liver and gastrointestinal tract) and maybe confirmed using biopsies. Therefore, at the time of diagnosis, patients can already have substantial organdamage. Recently, there has been a push to develop multi-biomarker immunoassays for conclusive acute GVHDdiagnosis before the onset of symptoms, because in the event that a patient develops acute GVHD, it is criticallyimportant to treat them early to prevent organ damage8,9,14–16. Unfortunately, as explained above, it is difficult to

OPEN

SUBJECT AREAS:BIOMARKERS

POLYMER CHEMISTRY

ASSAY SYSTEMS

Received10 January 2014

Accepted16 April 2014

Published2 May 2014

Correspondence andrequests for materials

should be addressed toS.P. (sophpacz@iu.

edu) or S.T.(takayama@umich.

edu)

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 1

develop and implement multi-biomarker panels for clinical settingsbecause of cross-reactions among antibodies that complicate themultiplexed validation of new biomarkers due to false readouts.Misdiagnosis of acute GVHD can be particularly dangerous topatients since immunosuppressive treatments themselves can resultin sepsis and early malignancy relapse via loss of the graft-versus-tumor effect.

Many multiplex assays rely on the application of a cocktail ofdetection antibodies. After extensive optimization, such cocktailscan often provide satisfactory results17–21. However, this type ofoptimization can be costly and time consuming. Moreover, if addi-tional biomarkers are added to the panel or the antibody reagentschange, the process of optimization must be repeated. To avoid thisproblem, we previously resorted to performing serial singleplexELISAs22.

Previously, several other approaches were developed to mitigateantibody cross-reactions for improved multiplex protein detec-tion23–25. For example, antibody colocalization microarrays usealigned spots of capture and detection antibodies that are dispensedin microliter volumes in air on nitrocellulose membranes. However,there is a tendency for the antibody solutions to evaporate underambient conditions, leading to antibody and biomarker degradation.Bead-based assays, such as LuminexTM, can theoretically achieve upto 100-plex. In reality, however, capture and detection antibodycross-reactions limit LuminexTM-based protein detection to the ini-tial screening of biomarkers, rather than clinical biomarker verifica-tion26. Finally, advanced array printing technologies, such ascontinuous flow print heads27,28, allow aligned deposition of multipleprotein solutions (including antibodies and sample solutions) withlimited evaporation, but require specialized, non-standard labor-atory equipment that may be difficult to access by laboratories andclinics.

Here, we eliminate the problem of detection antibody cross-reac-tions in multiplex ELISA by using aqueous two-phase systems(ATPSs) composed of the phase-separation-promoting polymerspolyethylene glycol (PEG) and dextran (DEX)27,28 to confine detec-tion antibody solutions in a fully aqueous environment to regionswhere complementary capture antibodies are immobilized29,30. Thismethod, referred to as ATPS-ELISA, works on three principles: i.droplets of the denser DEX solution sink in the PEG solution andremain in contact with the assay plate during incubation; ii. inter-facial tensions between DEX-PEG and DEX-assay plate cause the

DEX droplets to form domes that remain in place; and iii. detectionantibodies are retained, without diffusive dispersion, in the DEXphase due to partitioning effects. We demonstrate the translationalpotential of this technology by simultaneously detecting at least fourdifferent antigens that were previously demonstrated to be associatedwith acute GVHD, a disease that requires multiple biomarker ana-lysis for definitive diagnosis and prognosis14. Our assay is not onlyfree from antibody cross-reactions, but also requires less antibodyand patient sample than conventional single biomarker ELISA.

ResultsDetection Antibody Micro-Droplet Patterning Prevents Cross-Reactions. ATPS-ELISA is performed using the same steps as con-ventional sandwich ELISA, except that ATPS-ELISA uses detectionantibody solutions that are deposited in DEX droplets over thecapture antibody spots (Figure 1 a and b) by micropipetting. Thisprevents cross-reactions between unwanted pairs of polyclonalcapture and detection antibodies, as well as from non-target anti-gen recognition by detection antibodies. The capture antibodies anddetection antibodies can be easily aligned using polystyrene plateswith embossed features (Figure S1), although it is also possible toperform the assay using planar polystyrene substrates, as shown inFigures 1 and 2. Figure 1 c shows the effectiveness of ATPS-ELISA ateliminating antibody cross-reactions. We intentionally spottedpolyclonal antibodies that are more prone to cross-react. One spotcontained an immobilized anti-human ST2 capture antibody, whilethe adjacent spot contained an immobilized anti-goat antibody thatrecognized the detection antibody (a goat anti-human ST2 anti-body). After incubating with a solution containing ST2, the goatdetection antibody was either deposited above the mouse captureantibody-antigen complex in DEX using an ATPS or bath appliedin the conventional manner without using ATPS (Figure 1 c, d).There was no cross-reactive signal for the ATPS format (Figure 1c), demonstrating that the detection antibodies remained confined toDEX and were unable to diffuse away to react with the anti-goatantibody spot. On the other hand, conventional bath applicationallowed detection antibodies to freely circulate, resulting in strongsignals at both the appropriate capture antibody spot and the spotcontaining the anti-goat antibody (Figure 1 d). This illustrates thepossibility of false-positive readouts that may lead to misinterpre-tation of the biomarker panel when inappropriate combinations ofdetection antibodies are used in conventional multiplex systems.

Figure 1 | ATPS-ELISA prevents polyclonal detection antibody cross-reactions. Multiplex ATPS-ELISA (a) and conventional sandwich ELISA

(b) share similar procedures, shown in steps i., ii., and iii. However, by co-localizing detection antibodies (dashed line antibody symbols) in the DEX

phase over the corresponding capture antibodies (solid line antibody symbols) through the use of simple micropipetting onto custom plates with features

designed for micropipette tip alignment, ATPS-ELISA produces signals without any possibility of cross-reactions. (c) Goat detection antibodies

were not captured by a neighboring spot coated with an anti-goat capture antibody, indicating that antibodies did not diffuse out of the DEX droplet to

cause cross-reactions. (d) Bath application of detection antibodies resulted in cross-reactions at the neighboring capture antibody spot.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 2

As shown in Figure 2 a, phycoerythrin (PE)-IgG was well retainedwithin a DEX droplet over the course of 2 h. Furthermore, analysis ofthe distributions (i.e., the partition coefficients) of the detection anti-bodies within equilibrated ATPSs indicated that the antibodies par-titioned predominantly to DEX (Figure 2 b). The cat eye-shapedsignal in Figure 2 c shows that ATPS-ELISA can localize the detectionantibodies to specific regions and that signal is restricted only to theregions where both the detection antibody solution droplet andthe surface-adsorbed capture antibody spot overlap. In Figure 2 d,the detection antibody was bath applied, producing a larger circularsignal spot because the detection antibody solution covered the entirecapture antibody spot. These results demonstrate the ability ofATPS-ELISA to suppress detection antibody cross-reactions, whilehighlighting the need to align the detection antibody solution drop-lets over the capture antibody spots to avoid signal reducing artifacts.

The impact of optical crosstalk was not significant in our system,as assessed by signal area intensity plots and background measure-ments (Figure S2). These plots also demonstrated that the signalintensities of the sample regions were relatively uniform for bothhigh and low antigen concentrations using ATPS detection antibodyspotting.

ATPS-ELISA Facilitates Multiplex GVHD Biomarker Detection.Custom plates molded from polystyrene were used to facilitatecapture antibody and detection antibody alignment in our mul-tiplex assays. The plates were designed to prevent movement of theDEX droplets after pipetting and to prevent inadvertent mis-alignment of the capture and detection antibody spots. Prior tofabricating our final plate, we tested a range of antibody welldiameters (0.5 mm to 2 mm) with different edge-to-edge spacing(0.3 mm to 0.6 mm) and a depth of ,110 mm. Using the smallestwell diameter and spacing, it was possible to deposit a 7 3 7 array

(49-plex) of detection antibodies, as well as 4 3 4 (16-plex) and 3 3 3(9-plex) detection antibody arrays (Figure 3). Figure 3 a, b shows anexample of a 9-plex array with an experimental setup similar toFigure 1 c, d. When applied in DEX, the goat anti-ST2 detectionantibodies correctly localize to the 5 spots in the 9-plex arraywhere the anti-ST2 capture antibodies are located with bound ST2protein (Figure 3 a). However, the greater number of anti-goatcapture antibody spots (4 spots as opposed to the single spots inFigure 1 d; i.e., approximately 4 times the area of spotted captureantibody) serve to sequester the goat-anti-ST2 detection antibodieswhen they are bath applied in a 10 mL volume, producing falsenegative readouts at the locations where the ST2 captureantibodies are localized and false positive readouts where the anti-goat antibodies are localized (Figure 3 b). Figure 3 c shows anexample of 16-plex ST2 detection antibody patterning for a bathapplied ST2 capture antibody. Bath application of the detectionantibody results in chemiluminescent signal throughout the entiresample well (Figure 3 d).

Based on the number of biomarkers selected for our study, wedesigned our final plate with four antibody wells (4-plex; Figure 4a) that were 1.5 mm in depth. These antibody wells functioned tohold the capture and detection antibody solutions in place duringplate manipulation and transport. They also provided visual andtactile cues to ensure that we could easily perfect the alignment ofthe capture and detection antibody solutions using a multi-pipettor.Standard curves for 4 acute GVHD biomarkers (HGF, elafin, ST2and TNFR1) were generated using the optimized antibody condi-tions suggested by the manufacturer for both multiplex ATPS-ELISAand individual ELISA. The limits of detection and linear dynamicrange (LDR) values for ATPS-ELISA were generally comparable toindividual sandwich ELISAs (Figure 4 b–f), although in some cases(e.g., ST2 and TNFR1), the standard curves for ATPS-ELISA reached

Figure 2 | DEX droplets assume dome shapes that change very little in size and shape over the course of the incubation period. (a) Antibodies are

retained in the DEX domes over this period as indicated by the overlap between FITC-DEX and PE-IgG. Scale bar 5 1 mm. (b) Biotin-labeled ELISA

detection antibodies partition favorable to the DEX phase. Partition coefficients were measured by blotting detection antibody fractions from PEG and

DEX on PVDF membranes and detecting the antibody levels by way of streptavidin-HRP chemiluminescence. Partitioning can be further improved by

modifying ATPS formulations. (c) Partial overlap of the capture and detection antibodies results in a cat eye shape that can only be produced if antibodies

are well retained in the DEX droplet. (d) Bath application of detection antibodies produced a circular signal area.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 3

saturation slightly before the standard curves for individual sand-wich ELISA. The LoD and LDR values were also acceptable foranalysis of acute GVHD patient samples.

Validation of the Multiplex GVHD Assay. We analyzed plasmasamples from three patient groups: healthy controls (n520),allogeneic bone marrow transplant patients who did not manifestsymptoms of acute GVHD (GVHD –; n519, median 28 days post-transplant) and allogeneic bone marrow transplant patients who hadbeen diagnosed with acute GVHD (GVHD1; n532, median 28 dayspost-transplant). The biomarkers and samples we used were usedpreviously as part of several larger studies that analyzed GVHDstatus using serial singleplex ELISA9,14,16,22. We were blind to theGVHD status of the samples during the experiment and the imagequantification. As expected, the GVHD1 patient group hadsignificantly higher levels of HGF, elafin, ST2 and TNFR1 than theGVHD – group and the healthy donor group (Figure 5 a–d),demonstrating the effectiveness of our system at measuringmultiple biomarkers from patient samples.

The ATPS-ELISA procedure, allowed us to accurately assess theGVHD status of the samples (Figure 5 e–h). For example, ATPS wasslightly more effective at discriminating elafin levels betweenGVHD2 and GVHD1 patient samples, because the higher levelof elafin background produced more inconsistent readings in wellplates compared to the low-profile ATPS-ELISA wells, possibly dueto differences in the plate geometries. For example, due to the lower

surface area of the capture antibody-coated regions in the ATPS-ELISA wells (,1.77 mm2), we would expect there to be greater signalper unit volume (10 mL) of antigen compared to the 384-well platereservoirs used for single ELISA (.23 mm2). This is reflected in ourstandard curve data, which show that the ATPS-ELISAs enter linearranges for signal detection at lower antigen concentrations than theconventional single ELISAs. These observations are consistent withthe observations of others that suggest that smaller spot sizes relativeto the assay volumes produce superior LoD and LDR values31,32. It isalso possible that ATPS-ELISA benefits from microscale surfacelocalization effects that enhance detection antibody binding, moreeffective washing due to the shallow well profiles and less depletion ofchemiluminescent substrate, all of which can affect LoD and LDR. Inaddition, we note that the elafin ELISA reagents tended to be lessrobust and expire more quickly than the reagents from the other kits,which may explain why the elafin standard curves displayed higherlimits of detection and lower linear dynamic ranges than the otherbiomarkers in the panel.

Bland-Altman analysis was used to evaluate the agreementbetween ATPS-ELISA and standard single ELISA (Figure 6). Weobserved proportional and magnitude biases between the ATPS-ELISA and single ELISA formats, indicating that there were discrep-ancies between the two methods. These discrepancies were generallytolerable at lower biomarker concentrations, for example, below2,500 pg/mL for HGF, 36,000 pg/mL for elafin, 4,700 pg/mL forST2 and 1,800 pg/mL forTNFR1 (i.e., concentrations that approx-imate the clinically important thresholds for GVHD diagnosis).Although a larger patient cohort would be required to determineclinical cutoffs for our assay, the approximate clinical cutoffs canbe obtained from previous biomarker studies for HGF33, elafin9,ST216 and TNFR114. It is important to note that although theBland-Altman analyses indicated discrepancies between the twoassay formats, this does not necessarily indicate that one format issuperior to the other. To provide a comparison of the performance ofthe two methods, we conducted receiver operating characteristic(ROC) analysis (Figure S3). In our hands, the ATPS format slightlyoutperformed the single ELISA format, as assessed by the areas underthe curves, indicating that the ATPS method might provide superiorsensitivity and specificity. Larger experiments involving greaternumbers of biomarkers and additional samples, along with experi-menter-to-experimenter comparisons, will be required to conclu-sively prove this assertion. However, these experiments are beyondthe scope of the present study, which focuses on the development of apromising new technology for detection antibody patterning.

It is also important to emphasize that the multiplex biomarkerpanel was critical for obtaining high sensitivity (correct classificationof patients with GVHD) and high specificity (correct classification ofpatients without the disease). For example, one of the GVHD–patients had high levels of HGF (6615 pg/mL). Based only on thelevels of this biomarker, this patient could have been diagnosed withacute GVHD and could have potentially received unnecessary treat-ment. However, with the inclusion of the three other markers mea-surements that were all low, this patient can be considered GVHD–and avoid unnecessary treatments that may cause dangerous sideeffects. In other cases, a single biomarker did not reach the diagnosticthreshold for GVHD in spite of the patient’s GVHD1 status. In thesecases the information obtained from the other three biomarkers inthe panel that are all elevated can be used to correctly assess thedisease status of the patient.

DiscussionWe demonstrated that our system can utilize chemiluminescence todetermine the levels of biomarkers without antibody cross-reactionsthat can cause false positive and false negative signals using heavilycross-reacting antibodies as a test example (Figures 1, 2 and 3). Wefurther validated the usefulness of the ATPS-ELISA technology by

Figure 3 | ATPS ELISA can be used to pattern 9-plex and 16-plex arraysof detection antibodies. (a) A 9-plex version of the experiment shown in

Figure 1 c that demonstrates how ATPS-ELISA suppresses cross-reactions.

Goat anti-ST2 detection antibody solutions are dispensed as ATPS

droplets only to the 5 regions with spotted ST2 capture antibodies and ST2

antigen. These 5 regions generated true-positive signals. The 4 other

regions spotted with anti-goat capture antibodies that would cross-react

and give false positive signals with the goat anti-ST2 detection antibody did

not receive any detection antibody solution droplets and thus resulted in

no signal, demonstrating suppression of false positive signals. (b) A 9-plex

version of the experiment shown in Figure 1 d. Bath applied goat anti-ST2

detection antibodies become sequestered by the 4 anti-goat antibody spots.

This produces 4 false positive readouts and interestingly 5 false negative

signals with very low (but detectable) signal levels because there is

insufficient ST2 detection antibody available to bind to the ST2 sandwich

regions despite the fact that the anti-ST2 capture antibodies are bound to

ST2. (c) Representative image of a 16-plex ATPS sandwich ELISA for ST2

with bath application of the capture antibody and localized dispensing of

detection antibody in DEX droplets. (d) Same experiment as in c but

without localized dispensing of detection antibody in the DEX droplets.

The result is a signal from the entire well rather than localized signals.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 4

multiplexing 4 GVHD biomarkers that previously had to be mea-sured using sequential single ELISA14,22. Although this is a relativelysmall panel, 4-plex represents a threshold at which running singleELISAs begins to become prohibitive in terms of assay time, materialcosts and sample consumption, especially when using precious sam-ples where serial single ELISAs must be performed14,22. Furthermore,the working principles of the assay ensures that the degree of plexingis not limited by increasing concentrations of antibody cocktail solu-tions, since each detection antibody reagent solution is confined bythe ATPS only to its corresponding capture antibody spot.

Our assay offers several advantages over conventional single bio-marker ELISAs including the use of small plasma volumes (,10 mLtotal of diluted plasma for 4 biomarkers and ,1 mL of patient sample

per well), cost savings (less antibody consumption) and use of readilyavailable plate materials and detection systems. For example, per-forming similar experiments using the individual wells of a conven-tional 384-well plate would require at least 40 mL of sample(,10 mL/well) to fill the 4 wells required for each of the 4-biomarkerstested. In contrast, our detection antibody patterning technologyuses only 10 mL of sample total, since the 4 antibody spots are pat-terned within a low profile well of 6.5 mm diameter (an intermediatediameter between 96 and 384 format wells). We used 384 well formatplates for our comparison to singleplex ELISA. Although the smallerwell size allowed us to use 10 mL sample volumes for both the ATPSand regular ELISA assay formats, we did observe higher backgroundand slightly worse ROC when using 384 well plates, likely due to a

Figure 4 | Multiplex ATPS-ELISA for GVHD biomarkers. (a) A 3D rendering showing the custom plate design consisting of 4 antibody insets

within a common shallow sample well. (b) Representative images of the chemiluminescent standards. The concentrations listed above each image are the

same for each biomarker in the panel. The images correspond to every third point on the quarter-logarithmic dilution curve from 10,000 pg/mL to zero.

In each of the images the spot at the top corresponds to HGF, the spot on the right corresponds to elafin, the spot at the bottom corresponds to ST2 and the

spot on the left corresponds to TNFR1. It is apparent from these images that elafin has the highest limit of detection due to its high background

(elafin can be detected in healthy plasma as well as in GVHD1 plasma). Since the biomarker standards contained a 10% healthy pooled plasma to adjust

for matrix affects (see ‘‘Plasma samples’’ in the Methods section), this background can be attributed to baselines levels of elafin and should not be confused

as an example of a false positive signal. (c–e) Standard curves for all four GVHD biomarkers in PBS generated by densitometric quantification of

chemiluminescence images compared to individual sandwich ELISAs (dashed lines). Error bars represent standard error of the mean.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 5

difference in the capture antibody area to sample volume ratio andpossibly also due to the deep well geometry of conventional 384 wellplates that can increase pipetting errors and make it more difficult towash completely.

The ATPS-ELISA also provides advantages over miniaturized sin-gle ELISA systems, such as Droparray, which use non-aqueous solu-tions (i.e., oils) to cover the aqueous droplets34. Droparray utilizesdroplet sizes (,2 mL) on the same order as the DEX droplets used inour ATPS-ELISA. However, Droparray is not designed as a multipleximmunoassay, since, in a similar fashion to a 384-well plate assay, thereagents, which include the blocking buffer, samples and antibodies,are all dispensed in separate droplets. Thus, Droparray is essentially aminiaturized version of singleplex ELISA, where each droplet regionis treated as a separate well. The strength of our assay is that it uses acommon sample (plasma) volume applied over all of the captureantibody spots in the panel, which minimizes the impacts of samplehandling and pipetting errors that may occur with repeated pipettinginto multiple separate wells.

In contrast to commercially available multiplex biomarker ana-lysis systems, such as LuminexTM, ATPS-ELISAs can be performedwith a broad range of detection antibodies, including polyclonalantibodies and cross-reacting antibodies. LuminexTM, on the otherhand, must be carefully tested when introducing new antibodyreagents, since it is not specifically designed to suppress antibodycross-reactivity. Moreover, our assay does not require specializedanalyzers or sample handling devices. Instead, our assays can be readon common readers using plates that can be easily fabricated in manylaboratories from bulk polystyrene by a hot-embossing-based rapidprototyping process. We envision such plates to become available forbroader dissemination by injection molding. Finally, ATPS-ELISAsare highly scalable. By depositing antibody reagents in microdro-plets, we prevent antibody cross-reactions and obtain multiplexedassays with comparable results to conventional singleplex ELISAs.

While many potential biomarkers have been discovered, there aremultiple barriers (especially for multiplex biomarker assays) to bio-marker verification, qualification and FDA approval. One technolo-gical barrier has been the lack of multiplex immunoassays that areefficient in usage of precious samples (consume less per run), robustin providing quantitative data without an increase of background orcross-reactions, rapidly customizable and free from reliance on dif-ficult to access reagents, plates and hardware32,35. In addition, poorantibody specificity is problematic for immunoassays in general.Even with the best antibodies, high background and loss of linearityoccurs with multiplexing due to higher total antibody concentra-tions. Our assay addresses these key areas. For example, the totalconcentration of detection antibodies in our system is low regardlessof the degree of multiplexing, since all of the antibody solution drop-lets are separately spotted.

We also remove the bottleneck for clinical translation in severalways. First, multiplex signals become more robust due to lack ofcross-reactions. Second, the validation procedure becomes easierbecause each biomarker in the panel is arrayed (or patterned) inde-pendently of the others. Therefore, if reagents change or one assaydoes not work, the remaining biomarkers are still valid. Finally, theincreased variety of antibodies that can be used with ATPS-ELISAreduces the time and cost associated with the development andvalidation of new panels of multiplex immunoassays.

In the context of acute GVHD diagnosis and biomarker qualifica-tion, it is important to use a panel of multiple biomarkers14.Corticosteroids, which are general immunosuppressants, are usedas the first line of treatment for patients presenting with GVHDsymptoms, but fewer than half of patients treated with this standardtherapeutic regimen have a sustained response36. Early identificationof patients who will not respond to GVHD therapy is importantbecause patients who develop steroid refractory acute GVHD havepoor prognosis with an expected 1 year mortality of up to 90%37.

Figure 5 | Multiplex ATPS-ELISA enables robust detection of GVHD biomarkers in patient plasma samples. The ATPS-ELISA multiplex detection

system was used to probe human plasma for four biomarkers: (a) HGF, (b) elafin, (c) ST2 and (d) TNFR1. The GVHD1 group displayed significantly

higher levels of all four biomarkers compared to the GVHD – and healthy control groups (p,0.05 by one-way ANOVA with Dunn’s multiple comparison

test). Measurements of (e) HGF, (f) elafin, (g) ST2 and (h) TNFR1 from patient plasma using individual sandwich ELISAs are provided for comparisons.

Significance between the GVHD1 group and the GVHD- and healthy control groups was obtained for HGF and TNFR1 (p,0.05 by one-way ANOVA

with Dunn’s multiple comparison test) for individual ELISAs, but not for elafin or ST2, although ST2 values for the GVHD1 group tended to be higher

than the other groups. Error bars represent standard error of the mean.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 6

Improving the risk-stratification of patients with GVHD using bio-marker panels may permit early evaluation of additional therapiesbefore the development of treatment-resistant disease. Equallyimportant is the identification of patients who will respond well totreatment, which could allow for more rapid tapering of steroid regi-mens, thereby reducing toxicity such as increased risk for fatal infec-tions or increased incidence of relapses due to impaired graft versusleukemia in low risk patients. Therefore, it is important to avoid falsepositive diagnoses that result from immunoassay cross-reactions.

The ability to identify high-risk patients using biomarker panelsearly in their transplant course, before the development of GVHD,may permit more stringent monitoring and preemptive interven-tions. Another avenue for using these biomarker panels is post-trans-plant monitoring of the response once the treatment has started8.Patients should be regularly monitored following bone marrowtransplantation using non-invasive, rapid and standardized biomar-ker tests over the first month post-transplant, at the onset of acuteGVHD clinical symptoms and throughout the GVHD treatment.

The compatibility of ATPS-ELISA with plasma suggests that ourassay will also be compatible with other bio-fluids, such as urine,cerebral spinal fluid, salivary fluid and cell lysates. In addition, ourtest uses less sample, which is important for large retrospective stud-ies where the volume of samples are often limited due to limitedstorage space. The judicious use of patient samples will also be rel-evant to pediatrics, were the volumes of samples are limited due tothe size of children. Further studies with more rigorous standards for

assay validation will enable translation to clinical biomarker quali-fication and determination of the clinical cutoffs for our assay38–40.

In conclusion, ATPS-ELISA provides multiplex biomarker detec-tion for validation of new biomarker panels and diagnosis of complexdiseases, such as GVHD, where custom multiplex immunoassays arerequired. The ATPS-ELISA technology prevents cross-reactionsamong antibody reagents, reducing the risk of false positive and falsenegative detection of disease markers. This strategy is compatiblewith standard immunoassay reagents, workflows and equipment,allowing it to be easily implemented in clinical and laboratorysettings.

MethodsReagents. ATPSs were created from aqueous solutions of 20 wt% DEX 500,000(Pharmacosmos) and 20 wt% PEG 35,000. ELISA kits for human sTNFR1 (cat #DY225), human HGF (cat # DY294), human ST2 (cat # DY523) and human trappin-2/elafin (cat # DY1747) were purchased from R&D Systems. SuperSignal ELISAFemto Substrate (cat # 37074) was purchased from Thermo Scientific. Identicalreagents, concentrations and incubation times were used for both the ATPS-ELISAand single ELISAs according to the manufacturer’s recommendations.

Partition Coefficients. The partition coefficients for the various biotinylateddetection antibodies were determined by dot blot on polyvinylidene fluoride (PVDF)membranes. Briefly, the antibodies were diluted 15500 and thoroughly mixed inATPSs consisting of 10% PEG 35,000 and 10% DEX 500,000. The tubes were thencentrifuged at 400 rcf for 15 min to cause complete phase separation of the PEG andDEX. The PEG and DEX fractions were collected and placed in separate tubes at 4uC.The PVDF membranes were prepared by soaking in methanol for 15 s, followed bysoaking in water for 1 min and PBS for 5 min. The PVDF membranes were

Figure 6 | Bland-Altman analysis for (a) HGF, (b) elafin, (c) ST2 and (d) TNFR1. The dashed horizontal lines represent the 2SD confidence

intervals and the solid horizontal line represents the mean difference between assay formats. We observed discrepancies between the two assay formats,

particularly at the higher biomarker concentrations.

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 7

transferred onto PBS-moistened chromatography paper and spots containing 0.5 mLof the separated PEG and DEX samples were pipetted directly onto the membranes.The membranes were then blocked in 1% BSA for 1 h and washed 4 times in PBScontaining 0.05% Tween-20. The membranes containing the PEG- and DEX-antibody spots were incubated in streptavidin-conjugated HRP for 1 h and washed 4times with PBS. Chemiluminescence signals were developed using SuperSignal ELISAFemto Maximum Sensitivity Substrate (37074, Thermo Scientific) and detected usinga Fluorchem M imager (Protein Simple).

ELISA Protocol. Polystyrene plates were spotted with capture antibodiesreconstituted in PBS (4 mL per spot). The plates were stored at 4uC overnight in sealedcontainers containing PBS-soaked Kimwipes to limit evaporation. The next day, theplates were washed five times with wash buffer (PBS, containing 0.05% Tween-20)and blocked with 3% casein for 1 h. After blocking, the plates were washed five times.The rapid and thorough washing performed in this step (and later after the detectionantibodies were applied) ensured that the antibodies in solution would not be able toproduce measurable cross-reactions with neighboring antibody spots. Next, the plateswere incubated with the sample solutions (10 mL each for ATPS-ELISA) for 2 h.Following antigen application, plates were washed five times and incubated for 2 hwith detection antibodies in either the traditional ELISA or ATPS-ELISA format. Forconventional singleplex sandwich ELISA, detection antibodies were bath appliedaccording to the manufacturer’s specifications. For ATPS-ELISA, the plate was firstfilled with a solution of PBS containing PEG and 0.1% casein. DEX dropletscontaining the appropriate biotinylated detection antibodies (4 mL in volume) werethen pipetted into the PEG over the corresponding capture antibody spots. Inaddition to providing phase-separation and confinement of the DEX droplets, therelatively large volume of PEG used to cover the plate prevented evaporation of thedetection antibody solutions. The 4-plex antibody solutions were applied using aMatrix adjustable multipipettor (Thermo). An Ultimus I pneumatic pump (NordsonEFD) connected to a glass capillary needle containing the DEX/antibody solution wasused to fill smaller wells (e.g., 9-plex assays or 16-plex assays). The details of thissystem and its operation are described elsewhere41,42. The plates were washed eighttimes followed by incubation with streptavidin-conjugated horseradish peroxidasefor 1 h. Thorough and rapid wash buffer replacement ensured that there was noopportunity for detection antibody exposure to adjacent spots during washing. Plateswere then washed five times and incubated with SuperSignal ELISA Femto MaximumSensitivity Substrate (37074, Thermo Scientific). Chemiluminescent signal wasdetected using either a FluorChem M imaging system or a Synergy Neo HTS Multi-Mode Microplate Reader (Biotek). For single ELISAs the manufacturer’s procedurewas followed, with the exception that 384-well plates (rather than 96 well plates) wereused with 10 mL assay volumes to conserve reagents.

Plate fabrication. To optimize the plate design and demonstrate the capabilities ofour multiplex system we fabricated a polystyrene ELISA plate using PDMS hotembossing. Xurography (razor writing) of ,110 mm-thick vinyl tape was used togenerate tape templates for the larger sample wells and the smaller detection antibodywells43. The tape regions corresponding to the sample and antibody well regions werediscarded and the sample well template was adhered to the antibody well template.The resulting double layer tape template was then adhered to a polystyrene dish.PDMS was cured above the tape template to form a reverse replica of the tape patternthat was used to emboss polystyrene to form the ELISA plate44. The final aluminumplate mold used to generate the plates for assay validation with the GVHD biomarkerswas designed in AutoCAD (Autodesk) and SolidWorks (Dassault Systemes). Analuminum mold containing a 9 3 9 array of 6.5 mm diameter sample wells that eachcontained 4 1.5 mm diameter antibody wells was fabricated using precision CNCmicromachining (Protomatic, Inc., Dexter, MI). Polystyrene dishes were heatedabove their glass transition temperature and the aluminum mold was stamped intothe dish to mold the plate to its final dimensions. The 1.5 mm-deep antibody wellsserved three purposes. First, they held the capture antibody spots in place in the eventof any unexpected plate movements during transport. Second, they provided a surfacefeature that could be detected visually and by contact with the pipette tips to align theDEX droplets containing the detection antibodies with the capture antibody patterns.Third, they prevented sliding of the DEX/detection antibody droplets during platetransport and incubation.

Plasma samples. Heparinized plasma samples were collected from patients whoreceived allogeneic bone marrow transplantation at the University of Michiganbetween 2000 and 2010. Plasma samples were collected under protocols approved bythe University of Michigan Institutional Review Board and stored at the University ofMichigan. The protein standard buffer consisted of PBS containing 10% healthypooled plasma and 1% FBS, while sample buffer was PBS containing 1% FBS. Initialspike-and-recovery experiments indicated that signal intensities varied depending onpatient plasma buffer and the sample matrix (plasma). To control for matrix effects,FBS was added to the PBS sample buffer such that appropriate signal intensities wereachieve when control plasma was spiked with a known concentration of recombinantbiomarker protein.

Statistical analysis. All plots and statistical analyses were carried out in Sigmaplotwith Sigmastat (Systat Software). Standard curves were generated over a wide range ofconcentrations to ensure that there were no hooking effects at high concentrationsand to assess the level of non-specific background signal at low concentrations. Thevalues for single ELISA (collected using a chemiluminescence plate reader) were

scaled to produce values of the same order of magnitude as the ATPS-ELISA values(collected using the Fluorochem M western reader). The standard curves were fitusing a four parameter logistic function. The limit of detection (LoD) was determinedfrom the equation LoD 5 LoB 1 1.645 (SDlow concentration sample), where SD isthe standard deviation and LoB is the limit of blank45. LoB was calculated from LoB 5

meanblank 1 1.645 (SDblank). Linear dynamic range (LDR) was determined usingLDR 5 maximum linear response/LoD.

1. Engvall, E. & Perlmann, P. Enzyme-linked immunosorbent assay (ELISA).Quantitative assay of immunoglobulin G. Immunochemistry 8, 871–874 (1971).

2. Belanger, L., Sylvestre, C. & Dufour, D. Enzyme-linked immunoassay for alpha-fetoprotein by competitive and sandwich procedures. Clin. Chim. Acta 48, 15–18(1973).

3. Crowther, J. R. Methods in Molecular Biology [The ELISA guidebook, Volume 149](Humana Press, New York, 2000).

4. Halbert, S. P. et al. Quantitative estimation by a standardized enzyme-linkedimmunosorbent assay of human T-cell lymphotropic virus type I antibodies inadult T-cell leukemia and acquired immune deficiency syndrome. J. Clin.Microbiol. 23, 212–216 (1986).

5. Voller, A., Huldt, G., Thors, C. & Engvall, E. New serological test for malariaantibodies. Brit. Med. J. 1, 659–661 (1975).

6. Paul, B., Dhir, R., Landsittel, D., Hitchens, M. R. & Getzenberg, R. H. Detection ofprostate cancer with a blood-based assay for early prostate cancer antigen. CancerRes. 65, 4097–4100 (2005).

7. Kim, J. H. et al. Osteopontin as a potential diagnostic biomarker for ovariancancer. JAMA-J. Am. Med. Assoc. 287, 1671–1679 (2002).

8. Levine, J. E. et al. Acute graft-versus-host disease biomarkers measured duringtherapy can predict treatment outcomes: a Blood and Marrow Transplant ClinicalTrials Network study. Blood 119, 3854–3860 (2012).

9. Paczesny, S. et al. Elafin is a biomarker of graft-versus-host disease of the skin. Sci.Transl. Med. 2, 13ra12 (2010).

10. Moody, M. D., Van Arsdell, S. W., Murphy, K. P., Orencole, S. F. & Burns, C.Array-based ELISAs for high-throughput analysis of human cytokines.Biotechniques 31, 186–190, 192–184 (2001).

11. Wiese, R., Belosludtsev, Y., Powdrill, T., Thompson, P. & Hogan, M. Simultaneousmultianalyte ELISA performed on a microarray platform. Clin. Chem. 47,1451–1457 (2001).

12. MacBeath, G. Protein microarrays and proteomics. Nat. Genet. 32 Suppl, 526–532(2002).

13. Ellington, A. A., Kullo, I. J., Bailey, K. R. & Klee, G. G. Measurement and qualitycontrol issues in multiplex protein assays: a case study. Clin. Chem. 55, 1092–1099(2009).

14. Paczesny, S. et al. A biomarker panel for acute graft-versus-host disease. Blood113, 273–278 (2009).

15. Paczesny, S., Choi, S. W. & Ferrara, J. L. Acute graft-versus-host disease: newtreatment strategies. Curr. Opin. Hematol. 16, 427–436 (2009).

16. Vander Lugt, M. T. et al. ST2 as a Marker for Risk of Therapy-Resistant Graft-versus-Host Disease and Death. New Engl. J. Med. 369, 529–539 (2013).

17. Charlermroj, R. et al. Comparison of techniques to screen and characterizebacteria-specific hybridomas for high-quality monoclonal antibodies selection.Anal. Biochem. 421, 26–36 (2012).

18. Gehring, A. G., Albin, D. M., Reed, S. A., Tu, S. I. & Brewster, J. D. An antibodymicroarray, in multiwell plate format, for multiplex screening of foodbornepathogenic bacteria and biomolecules. Anal. Bioanal. Chem. 391, 497–506 (2008).

19. Karoonuthaisiri, N. et al. Development of antibody array for simultaneousdetection of foodborne pathogens. Biosens Bioelectron. 24, 1641–1648 (2009).

20. Huang, R. P., Huang, R., Fan, Y. & Lin, Y. Simultaneous detection of multiplecytokines from conditioned media and patient’s sera by an antibody-basedprotein array system. Anal. Biochem. 294, 55–62 (2001).

21. Wang, D., Liu, S., Trummer, B. J., Deng, C. & Wang, A. Carbohydrate microarraysfor the recognition of cross-reactive molecular markers of microbes and host cells.Nat. Biotechnol. 20, 275–281 (2002).

22. Fiema, B. et al. High Throughput Sequential ELISA for Validation of Biomarkersof Acute Graft-Versus-Host Disease. J. Vis. Exp. e4247 (2012).

23. Pla-Roca, M. et al. Antibody colocalization microarray: a scalable technology formultiplex protein analysis in complex samples. Mol. Cell. Proteomics 11,M111.011460 (2012).

24. Fredriksson, S. et al. Protein detection using proximity-dependent DNA ligationassays. Nat. Biotechnol. 20, 473–477 (2002).

25. Ekins, R. P. Multi-analyte immunoassay. J. Pharmaceut. Biomed. 7, 155–168(1989).

26. Djoba Siawaya, J. F. et al. An evaluation of commercial fluorescent bead-basedluminex cytokine assays. PLoS One 3, e2535 (2008).

27. Eddings, M. A. et al. Improved continuous-flow print head for micro-arraydeposition. Anal. Biochem. 382, 55–59 (2008).

28. Natarajan, S., Hatch, A., Myszka, D. G. & Gale, B. K. Optimal conditions forprotein array deposition using continuous flow. Anal. Chem. 80, 8561–8567(2008).

29. Tavana, H. et al. Nanolitre liquid patterning in aqueous environments for spatiallydefined reagent delivery to mammalian cells. Nat. Mater. 8, 736–741 (2009).

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 8

30. Albertsson, P. A. Partition of cell particles and macromolecules: separation andpurification of biomolecules, cell organelles, membranes, and cells in aqueouspolymer two-phase systems and their use in biochemical analysis and biotechnology[3rd ed.] (Wiley, New York, 1986).

31. Hartmann, M., Roeraade, J., Stoll, D., Templin, M. F. & Joos, T. O. Proteinmicroarrays for diagnostic assays. Anal. Bioanal. Chem. 393, 1407–1416 (2009).

32. HTStec. ELISA Trends 2010. Date of Access: January 9, 2014.http://www.htstec.com/consultancyitem.aspx?Item5297.

33. Harris, A. C. et al. Plasma biomarkers of lower gastrointestinal and liver acuteGVHD. Blood 119, 2960–2963 (2009).

34. Kozak, K. R. et al. Micro-volume wall-less immunoassays using patterned planarplates. Lab. Chip 13, 1342–1350 (2013).

35. Master, S. R., Bierl, C. & Kricka, L. J. Diagnostic challenges for multiplexed proteinmicroarrays. Drug Discov. Today 11, 1007–1011 (2006).

36. MacMillan, M. L. et al. Response of 443 patients to steroids as primary therapy foracute graft-versus-host disease: comparison of grading systems. Biol. BloodMarrow Tr. 8, 387–394 (2002).

37. Arai, S., Margolis, J., Zahurak, M., Anders, V. & Vogelsang, G. B. Poor outcome insteroid-refractory graft-versus-host disease with antithymocyte globulintreatment. Biol. Blood Marrow Tr. 8, 155–160 (2002).

38. Boja, E. S. et al. The journey to regulation of protein-based multiplex quantitativeassays. Clin. Chem. 57, 560–567 (2011).

39. Boja, E. S. & Rodriguez, H. Regulatory considerations for clinical massspectrometry: multiple reaction monitoring. Clin. Lab. Med. 31, 443–453 (2011).

40. Kaiser, J. Clinical medicine. Biomarker tests need closer scrutiny, IOM concludes.Science 335, 1554 (2012).

41. Frampton, J. P. et al. Microbubbles: Aqueous Two-Phase System Patterning ofMicrobubbles: Localized Induction of Apoptosis in Sonoporated Cells. Adv.Funct. Mater. 23, 3366–3366 (2013).

42. Frampton, J. P., Shi, H., Kao, A., Parent, J. M. & Takayama, S. Delivery of Proteasesin Aqueous Two-Phase Systems Enables Direct Purification of Stem Cell Coloniesfrom Feeder Cell Co-Cultures for Differentiation into FunctionalCardiomyocytes. Adv. Healthcare Mater. 2, 1440–1444 (2013).

43. Bartholomeusz, D. A., Boutte, R. W. & Andrade, J. D. Xurography: Rapidprototyping of microstructures using a cutting plotter. J. Microelectromech. S14,1364–1374 (2005).

44. Goral, V. N., Hsieh, Y. C., Petzold, O. N., Faris, R. A. & Yuen, P. K. Hot embossingof plastic microfluidic devices using poly(dimethylsiloxane) molds. J. Micromech.Microeng. 21 (2011).

45. Armbruster, D. A. & Pry, T. Limit of blank, limit of detection and limit ofquantitation. Clin. Biochem Rev. 29 Suppl 1, S49–52 (2008).

AcknowledgmentsThe authors wish to acknowledge funding from the following agencies: ST CoulterFoundation & NSF IIP1243080; SP National Institutes of Health grant RC1HL101102, EricHartwell Research fund and the Amy Strelzer Manasevit Research Program; JBW NSF DGE0718128 (ID: 2010101926).

Author contributionsJ.P.F. performed experiments, analyzed data, prepared the figures and wrote themanuscript. J.B.W. performed experiments, assisted with data analysis and figurepreparation and helped write the manuscript. A.B.S. assisted with experiments and helpedwrite the manuscript. M.T. assisted with experiments. S.P. and S.T. oversaw the project andthe preparation of the manuscript. All authors reviewed the manuscript.

Additional informationSupplementary information accompanies this paper at http://www.nature.com/scientificreports

Competing financial interests: The authors declare the following competing financialinterests. S.P. is author on the patent entitled ‘‘Methods for detecting graft versus hostdisease’’, US-61/542, 630. S.P. is a consultant for Viracor. J.B.W., A.B.S. and S.T. own stockfor PhasiQ, a company working on related technology. J.B.W., A.B.S., J.P.F. and S.T. areauthors of the patent ‘‘Systems and methods for multiplex solution assays’’, US-13/918494.

How to cite this article: Frampton, J.P. et al. Aqueous two-phase system patterning ofdetection antibody solutions for cross-reaction-free multiplex ELISA. Sci. Rep. 4, 4878;DOI:10.1038/srep04878 (2014).

This work is licensed under a Creative Commons Attribution 3.0 Unported License.The images in this article are included in the article’s Creative Commons license,unless indicated otherwise in the image credit; if the image is not included underthe Creative Commons license, users will need to obtain permission from the licenseholder in order to reproduce the image. To view a copy of this license, visithttp://creativecommons.org/licenses/by/3.0/

www.nature.com/scientificreports

SCIENTIFIC REPORTS | 4 : 4878 | DOI: 10.1038/srep04878 9